The drum index is a key indicator for evaluating pellets quality and is also the core regulatory object in the production process. To address the issues of strong lag in traditional detection methods and poor control effectiveness,as well as to achieve real-time and accurate forecasting of the quality of pellets, an ivy league optimization algorithm (MIVYA_DE) that improves upon the conventional algorithm by incorporating a combination of Latin Hypercube Sampling and Differential Evolution strategies is proposed. This algorithm addresses such shortcomings as uneven initial population distribution and weak local search later on,introducing latin hypercube sampling (LHS) and differential evolution (DE) strategies,and effectively enhancing global optimization capability. Modeling and validation results based on field data show that the proposed MIVYA_DE algorithm significantly improves prediction accuracy;Among them,the gradient improvement decision tree (GBDT) model coupled to this algorithm has the highest best-of-fit,with a determination coefficient (R² ) of 0. 948 7. This study verified the engineering practicality and promotion value of the improved algorithm.